Here below, referring to FIG. 1, we describe the prior art as regards image prediction in the field of the encoding or decoding of image sequences using the encoding of motion vectors for blocks of an image sub-divided into macroblocks which are then sub-divided into blocks. A block consists of a set of dots or pixels, where the block can have a square, rectangular, triangular or any other geometrical shape.
Thus, according to the H.264 technique for example, the encoder, for a current block to be encoded, computes a motion residual, also called a motion residue, by subtracting a predictor p from the current motion vector mv:εmv=mv−p, 
with εmv being the value of the motion information to be encoded, called a residue, my being the motion vector coming from motion estimation (“M Estimation” in FIG. 1) selected by the encoder and p being the computed predictor. The predictor corresponds for example to a modified median value of the motion vectors of the neighboring blocks of the current block considered. This residue is sent in an entropic encoder (for example of the “Context-adaptive Huffman variable-length coding” or CAVLC type) or an arithmetic encoder (for example of the “Context-adaptive binary arithmetic coding” or CABAC type).
Although the encoders present in H.264 are context-adaptive, i.e. although the current residue can be encoded differently as a function of the neighboring vectors (the code words are different), they do not make it possible to consider a goal of low-cost encoding of the motion information when computing this information.
To try and overcome this drawback, a prior-art technique described in the U.S. Pat. No. 6,178,265 seeks to compress the motion information to be encoded by encoding only the motion vectors selected as being relevant according to predetermined criteria and therefore reducing the number of motion information items to be encoded.
However, one drawback of this technique lies in the loss of information related to the selecting of the motion vectors to be encoded.
There is therefore a need for a novel technique for optimizing the encoding in terms of encoding tasks without lowering encoding performance.